The AI frontier is currently limited by physical power requirements.
The ability of operators to bring new capacity online to service compute is bound by a variety of regulatory and physical/market constraints.
Do folks not understand this?